Asset Management Registry Cost Per Case | Moring
Cost per case in asset management agent programs is the registry-attributed cost of an agent's model and tool consumption divided by the units of work it completes, giving operations and investment leaders a defensible unit economics measure for automating tasks such as reconciliation exceptions, client onboarding checks, and reporting cycles.
Short answer
Cost per case in asset management agent programs is the registry-attributed cost of an agent's model and tool consumption divided by the units of work it completes, giving operations and investment leaders a defensible unit economics measure for automating tasks such as reconciliation exceptions, client onboarding checks, and reporting cycles.
How it works
Cost per case is the unit economics measure that makes agent investment decisions rational. An asset management operations team can be told that an agent estate consumed a certain amount of model spend last quarter, but the number that changes decisions is what each completed exception review, onboarding check, or client report actually cost. The registry is the component that makes that number computable and trustworthy.
The computation has two halves. The numerator is consumption: tokens, tool invocations, and compute attributable to the agent during the period. The denominator is completed cases, which requires a definition of a case per workflow: one reconciliation exception investigated and dispositioned, one onboarding file processed, one client statement generated. Azure's agent economics guidance frames governance as the system that knows which agents exist and what they consume so AI can run as a managed investment; the registry supplies exactly that inventory.
Attribution through the registry matters because agents share platforms. Without the registry's linkage of agent to workflow and owner, spend aggregates at the platform level and unit costs cannot be computed. With it, every case's cost traces to the specific agent, the model version it called, and the accountable owner, which is the same lineage discipline that governs model use in the firm's SR 11-7 aligned framework.
Cost per case also exposes model selection economics. Where an agent's registry entry records the validated model version and consumption telemetry shows the cost of calls to it, teams can compare candidate models' cost per case against quality thresholds recorded in validation. This is model risk discipline applied to economics: a cheaper model is only admissible if the validated quality bar is met, and the registry carries both the version and the validation reference that make that check possible.
Operations leaders should treat cost per case as a managed metric with thresholds. Sweet Security's governance framework ties spend to ownership and enforceable limits; extending that pattern, the registry can hold a target cost per case per workflow and alert the named owner when actuals drift above it, whether from looping retries, prompt inflation, or a model version change that increased tokens per case.
The measure also disciplines build-versus-buy decisions. When internal agents' cost per case is known, comparing it to vendor pricing for the same workflow becomes evidence-based rather than rhetorical, and retirement decisions gain the same rigor: an agent whose cost per case exceeds the manual baseline it replaced, without compensating quality or capacity gains, is a documented candidate for rework or decommissioning.
For governance reporting, cost per case per workflow, joined to the registry's ownership and model lineage, gives technology and compliance leaders a single view where economic performance and control status appear together. That combination is what allows asset management firms to expand agent automation with confidence that each workflow earns its cost, under the same governance record that satisfies the firm's model risk and transparency obligations.
Frequently asked questions
What is cost per case for an AI agent in asset management?
Cost per case is the agent's attributed consumption, meaning tokens, tool calls, and compute for the period, divided by the completed units of work it produced, such as reconciliation exceptions dispositioned or client reports generated. The registry makes the measure computable by linking each agent to its workflow and owner, so every case's cost traces to the specific agent, model version, and accountable owner behind it.
Why does the registry matter for calculating cost per case?
Because agents share platforms, spend aggregates at the platform level without the registry's linkage and unit costs cannot be attributed to workflows. The registry supplies the join between consumption telemetry and the workflow, owner, and model version recorded per agent, which is the same lineage discipline asset management firms apply to models, so cost per case inherits the trustworthiness of the governance record rather than resting on estimated allocations.
How does cost per case inform model selection?
When the registry records the validated model version an agent calls and telemetry shows its cost per case, teams can compare candidate models' economics against the quality thresholds set in validation. A cheaper model is admissible only if it meets the validated quality bar, and the registry carries both the version and validation reference that make that comparison evidence-based rather than a trade-off made informally during development.
What should trigger review of an agent's cost per case?
Any drift of actual cost per case above the target recorded in the registry should alert the named owner for investigation, with common causes including looped retries, prompt inflation, or a model version change that increased tokens per case. Treating the target as a managed threshold, consistent with governance practice that ties spend to enforceable limits, keeps unit economics under continuous oversight instead of being examined only at budget reviews.